Link building is the practice of earning links from other websites to a page you want ranked. Links function as citations: a search engine treats a link from one site to another as a signal that the destination is worth reading. That model is 28 years old and still operating, but it no longer explains the whole picture. A second track now runs beside it, and the signals that predict visibility on that track are not links.
This guide covers what makes a link valuable, the methods used to acquire links, what Google’s link spam policy prohibits, and how to measure whether any of it worked.
Key Takeaways
- Branded web mentions correlate with AI Overview visibility at 0.664, against 0.218 for backlinks, across 75,000 brands studied by Ahrefs.
- Domain-level authority predicts which brand owns a topic in ChatGPT 52.5% of the time - statistically, a coin flip.
- Google’s Authority Score correlates with AI mentions at 0.65 but with AI share of voice at only 0.23, indicating a threshold effect rather than linear returns.
- Nofollow links show correlation with AI visibility close to follow links, which contradicts standard link-acquisition advice.
- Link acquisition still moves classic rankings. It is a weak lever on the citation track, where mentions do the work.
What Link Building Is
Link building is the process of acquiring hyperlinks from external websites to your own, for the purpose of increasing organic search visibility. The link is the unit. The acquiring is the practice.
A link carries three things a search engine can use: a signal of endorsement from the linking domain, a topical association created by the surrounding text, and a description of the destination in the anchor text. Links vary enormously in what they transmit. A contextual link inside the body of a relevant, well-trafficked article transmits far more than a footer link on an unrelated site.
The practice divides into two activities. Link earning produces links without a request, through content other people choose to cite. Link acquisition produces links through outreach, placement, or negotiation. Most programs run both. A strong marketing and sales strategy helps businesses build relationships, increase brand visibility, and create opportunities for organic growth.
Why Links Function as a Ranking Signal
Links function as a ranking signal because search engines have no other scalable way to measure whether the web considers a page credible. A page can claim authority. A link from an independent source is a third party spending its own reputation to say so.
PageRank and the Citation Model
PageRank models the web as a graph and treats every link as a vote weighted by the voting page’s own score. Google’s founding papers borrowed the model from academic citation analysis, where a paper’s influence is measured by how often other papers cite it.
The mechanism is recursive. A link from a page that itself has many strong inbound links transmits more than a link from an isolated page. This is why a single link from an established publication can outperform hundreds from low-quality sites - the graph position of the source determines the value of the vote.
Why Relevance Now Governs Authority
Relevance governs how much of a link’s authority actually reaches the destination. A link from a high-authority site in an unrelated subject area transmits less than a link from a moderate-authority site covering the same topic.
The reason is that modern retrieval systems evaluate links in context. The linking page’s subject matter, the surrounding paragraph, and the anchor text together tell the search engine what the destination is about. An off-topic link supplies an endorsement with no topical instruction attached, and it is treated accordingly.
The Two-Track Model: Rankings and AI Citations
Search visibility now runs on two tracks that respond to different signals. Track one is classic organic ranking, where links remain a primary input. Track two is citation in AI-generated answers, where brand mentions outperform links by a factor of roughly three.
Ahrefs measured this across 75,000 brands. Branded web mentions correlated with AI Overview visibility at 0.664. Backlinks correlated at 0.218. A December 2025 follow-up extending the analysis to ChatGPT and Google AI Mode found YouTube mentions correlating even higher, at approximately 0.737, with branded web mentions at 0.709 on AI Mode and branded anchor text at 0.527.
Correlation is not causation, and the direction of the relationship is genuinely ambiguous - brands that get mentioned often may simply be larger. The practical reading is narrower: link volume is a weak predictor of citation, so a program optimized entirely for link acquisition is optimizing for one track and leaving the other untouched.
Why Topic-Level Ownership Beats Domain Authority
Domain-level metrics barely predict which brand an AI assistant names. Semrush and Kevin Indig tracked 1,094 US categories monthly in ChatGPT from January through June 2026, with each category defined as a cluster of five buyer prompts covering definitions, comparisons, alternatives, use cases, and purchase questions.
Authority Score identified the category owner in 52.5% of head-to-head comparisons. Organic traffic managed 48.4%. Branded search volume, at 55.7%, was the only metric that registered above chance at all. Only 21% of the most-cited domains in a category were also the most-mentioned brands.
The consequence
The domains getting cited and the brands getting named are two different lists. A link program raises the first. Only editorial presence raises the second.
Ownership, once established, is stable. Brands holding a five-percentage-point lead retained first place in 90% of month-over-month comparisons. Narrow leads changed hands regularly.
What Makes a Link Valuable
A link’s value is determined by the relevance of its source, the authority of its source, its placement on the page, its anchor text, its link attribute, and the traffic and indexation status of the linking page. No single factor governs. A link scoring well on authority and badly on relevance transmits little.
The determinants below are the acquisition criteria. Each one is a question to answer before pursuing a link, not after.
Topical Relevance at Site, Page, and Section Level
Relevance operates at three levels, and all three matter. Site-level relevance asks whether the linking domain covers your subject area. Page-level relevance asks whether the specific linking page addresses a related topic. Section-level relevance asks whether the paragraph containing the link discusses something connected to the destination.
A link from a general news site inside an article about your industry, in a paragraph discussing your specific subject, scores well on all three. A link from an industry site in an unrelated roundup scores well on one.
Authority at Domain Level Versus Page Level
Domain authority describes the aggregate strength of a site. Page authority describes the strength of the specific page carrying the link. Page-level authority is the more direct predictor, because PageRank flows from pages, not from domains.
A link from a deep, un-linked page on a strong domain transmits less than the domain metric suggests. A link from a heavily-cited page on a mid-tier domain transmits more. Evaluating only the domain metric systematically misprices both.
Link Placement: Editorial Body, Author Bio, Footer, Sidebar
Editorial body links transmit the most value. Author bio links transmit less. Footer and sidebar links, which appear site-wide and were not placed by an editor for a reader, transmit least and can be discounted entirely.
The distinction reflects intent. A link placed inside a sentence because the writer wanted readers to follow it is an endorsement. A link appearing on every page of a site regardless of content is a template element.
Anchor Text Types and Distribution
Anchor text tells the search engine what the destination page is about. Six types appear in natural link profiles:
- Exact match - the target keyword verbatim
- Partial match - the keyword inside a longer phrase
- Branded - the company or product name
- Naked URL - the address itself
- Generic - “this article”, “here”, “read more”
- Image - the alt text of a linked image
Natural profiles skew heavily branded and generic. Profiles dominated by exact-match anchors are the clearest pattern signal of paid or manipulated placement, because editors writing for readers rarely choose a commercial keyword as anchor text.
Link Attributes: Dofollow, Nofollow, Sponsored, UGC
Attributes tell Google how to treat a link. Dofollow is the default and passes signals. Nofollow requests that the link not influence ranking. Sponsored marks paid placements. ugc marks user-generated content such as comments and forum posts.
Google has treated nofollow as a hint rather than a directive since March 2020, meaning the link may still be used for discovery and, at Google’s discretion, for ranking. On the citation track, the distinction matters even less: the Semrush and Kevin Indig backlink study found nofollow links showing correlation with AI mentions close to follow links, with the gap concentrated in the top authority tiers.
What this changes
A program that discards nofollow opportunities is discarding placements that perform comparably on the track where mentions matter most.
Traffic and Indexation of the Linking Page
A link on a page that receives no traffic and is not indexed transmits close to nothing. Indexation is the precondition - a page absent from the index cannot pass a signal it was never crawled to find.
Traffic is the secondary check. A page with real readers produces referral visits alongside the ranking signal, and referral traffic is the one link outcome that does not depend on any algorithmic interpretation.
Outbound Link Density on the Linking Page
The value transmitted by a page is divided among the links it carries. A page linking to eight destinations transmits more per link than a page linking to eighty.
High outbound density is also a quality signal in itself. Pages carrying hundreds of external links are usually directories, aggregators, or paid placement inventory rather than editorial content.
Link Neighborhood and Co-occurrence
The other sites a linking page points to describe the neighbourhood your link sits in. A page linking to recognised industry sources places you among them. A page linking to gambling, pharmaceutical, and unrelated commercial destinations places you among those.
Co-occurrence extends the same idea without a link. Appearing in a list, comparison, or article alongside trusted sources in your category creates an association that retrieval systems can use, whether or not any link is present.
INTERNAL LINK building topical authority → topical authority node
Link Velocity and Freshness
Link velocity is the rate at which new links appear. Sudden spikes disconnected from any event - no launch, no coverage, no viral content - form a pattern that automated systems detect.
Freshness matters independently. Recently acquired links from currently-updated pages signal ongoing relevance. A profile consisting entirely of links acquired years ago describes a site that was once notable.
Image Links Versus Text Links
Image links correlate with AI visibility more strongly than text links. The Semrush and Kevin Indig study measured image links at a Pearson coefficient of 0.415 against 0.334 for text links, with the effect more pronounced for higher-authority sources.
The likely mechanism is signal density. A linked image supplies alt text, filename, caption, and surrounding context together, where a text link supplies an anchor phrase alone. This inverts the standard assumption that image links underperform.
Geographic and Language Alignment
Do links from other countries count? Yes, and cross-border links from authoritative sources carry real value. Alignment still helps: a UK site targeting UK queries benefits more per link from UK sources, because geographic and language signals reinforce the relevance the link already carries.
Misalignment becomes a problem at scale. A profile where most links originate in a country you do not serve, in a language you do not publish in, describes a mismatch between your audience and your citations.
How Links Are Acquired
Links are acquired through nine broad methods, which differ in effort, realistic yield, risk exposure, and time to effect. The comparison below sets all nine against the same criteria.
This table compares each acquisition method by the effort it requires, the realistic number of links it produces per month of sustained work, its exposure to Google’s link spam policy, and how quickly links begin appearing.
Method | Effort | Monthly yield | Risk | Time to first links |
Digital PR and data assets | High | 5-40 | Low | 4-8 weeks |
Guest contribution | Medium | 2-8 | Medium | 3-6 weeks |
Broken link building | Medium | 1-5 | Low | 2-4 weeks |
Resource and directory placement | Low | 1-4 | Medium | 2-6 weeks |
Unlinked mention reclamation | Low | 2-10 | Low | 1-3 weeks |
Expert commentary | Medium | 3-12 | Low | 1-4 weeks |
Competitor link gap replication | Medium | 2-8 | Medium | 4-8 weeks |
Community participation | Medium | Indirect | Low | Ongoing |
Original research | High | 10-60 | Low | 6-12 weeks |
Yields assume one person working consistently on a site with an established topic. New sites should expect the lower bound or below.
Digital PR and Data-Led Assets
Digital PR earns links by producing something newsworthy and pitching it to journalists. The asset is usually original data, a survey, an index, or a tool, and the link arrives inside coverage rather than as the object of the request.
This method produces the highest-authority links available to most organizations, because national and trade publications do not accept placement requests but do accept stories. It carries the lowest guideline risk of any acquisition method, since the link is editorial by construction.
Guest Contribution
Guest contribution places an article you write on a publication that accepts external contributors, with a link to your site inside the article or author bio. The method works when the publication has genuine editorial standards and a real audience.
The risk sits in the degraded version. Google’s link spam policy names large-scale guest posting with keyword-rich anchor text as a violation. A contribution written for readers on a site that would publish it regardless of the link is fine. A template article placed on a site that exists to sell placements is not.
Broken Link Building
Broken link building finds dead links on relevant pages, then contacts the page owner suggesting your working resource as a replacement. The pitch succeeds because it identifies a defect the owner wants fixed.
Yield is low but conversion rates are higher than cold outreach, since the request carries value for the recipient before it carries value for you.
Resource and Directory Placement
Resource pages list useful references on a topic and accept submissions. Industry directories and professional association listings work the same way. Both produce modest, stable links.
The risk is in the tail. General-purpose directories with no editorial screening, no traffic, and no topical focus produce links that are at best worthless and at worst evidence of a manufactured profile.
Unlinked Mention Reclamation
Is an unlinked mention worth anything? Yes, and this is the reclassification the two-track model forces. An unlinked mention already contributes to the signal that predicts AI citation more strongly than backlinks. Converting it to a link adds ranking value without removing the mention value.
Find them by monitoring for your brand name, product names, and executive names across news, blogs, and forums. Contact the author, note the mention, and request the link. Response rates are high because the writer already chose to reference you.
Expert Commentary and Source Contribution
Journalists need sources. Responding to queries from reporters produces a quotation, an attribution, and usually a link, on publications that would never accept a placement request.
The method also generates the branded mentions that drive citation-track visibility, whether or not a link is included. A named quotation in a trade publication contributes to the mention signal even when the publication links nowhere.
Competitor Link Gap Replication
Link gap analysis identifies domains linking to competitors but not to you, then targets those domains. The logic is that a site willing to link to one company in a category may link to another.
The method’s limit is that it copies an existing profile rather than building a distinctive one. Links a competitor earned through original research or coverage cannot be replicated by asking.
Community and Forum Participation
Community participation builds presence in places where your audience already discusses the subject. Reddit, industry forums, and professional communities are increasingly cited directly in AI-generated answers.
Links from these sources carry ugc or nofollow attributes and pass limited ranking value. Their contribution is to the mention and co-occurrence signals, which is the track where they matter. Participation must be genuine - promotional posting is removed by moderators and produces nothing.
Link Earning Through Original Research
Original research earns links because it creates a fact that did not previously exist and that other writers need to cite. Surveys, experiments, aggregated datasets, and industry benchmarks all qualify.
The compounding property distinguishes this method from all others. A cited statistic continues attracting links for years after publication, and every citation is editorial by definition.
This is also the tactic with the strongest independent evidence behind it on the citation track. A study from Princeton, Georgia Tech, Allen Institute for AI, and IIT Delhi, presented at KDD 2024, tested optimization tactics across 10,000 queries and measured the effect of each on visibility in generative engines.
Adding statistics produced the largest measured gain. That result points back to original research as the acquisition method with the best return across both tracks.
What Google’s Link Spam Policy Prohibits
Google’s link spam policy prohibits links intended to manipulate ranking rather than inform readers. The policy covers buying and selling links for ranking purposes, excessive link exchanges, large-scale automated placement, and any link created for search engines rather than people.
The section below states what the policy says and what the observed consequences are. Judgement about acceptable risk belongs to the site owner.
Paid Links and Exchanges
Buying links that pass ranking signals violates the policy. Selling them does too. Paid placements are permitted when marked with the sponsored attribute, which tells Google not to count them for ranking - which is precisely why most paid link inventory is not marked.
Reciprocal linking becomes a violation at the point where it is excessive and driven by ranking rather than editorial reason. Two related businesses linking to each other once is not the same as a network of sites systematically exchanging links.
PBNs and Manufactured Networks
A private blog network is a set of sites created or acquired for the purpose of linking to a target site. Networks are prohibited under the policy regardless of the individual sites’ quality.
Detection targets shared patterns: registration data, hosting, template reuse, publication timing, content sourcing, and outbound link profiles. Networks are usually identified as networks rather than site by site, which means a single detection can devalue every link in the set at once.
Scaled Automated Placement
Automated tools that generate links across comment sections, forums, profiles, and low-quality sites produce links that are ignored at best. Google’s March 2024 spam policies update named scaled content abuse explicitly and expanded enforcement.
Scale is the detection signal. Thousands of links appearing across unrelated low-quality sites within a short window describes a pattern no editorial process produces.
Manual Actions Versus Algorithmic Devaluation
What happens if bad links point at your site? Two different things, with very different consequences. Algorithmic devaluation is the common outcome: Google ignores the links, and the site simply receives no benefit from them. Nothing appears in Search Console.
A manual action is issued by a human reviewer, appears in Search Console under Manual Actions, and suppresses rankings until resolved. Manual actions for unnatural links generally require evidence of a deliberate pattern rather than incidental low-quality links. Recovery requires cleanup and a reconsideration request, and takes weeks to months.
Google has stated repeatedly that most sites do not need to act on spammy inbound links, because the devaluation path handles them without penalty.
When Disavow Is and Isn’t Warranted
Disavow the links only when a manual action for unnatural links exists, or when a deliberate paid or manufactured link campaign is known to have run on the site. Those are the two warranted cases.
Preventive disavowal of ordinary low-quality inbound links is not warranted and carries its own risk: disavowing a domain removes any value it was passing, and the action is difficult to reason about after the fact. Sites without a manual action generally have nothing to gain.
Which Approach Fits Your Site’s Situation
The right approach depends on your site’s authority tier, age, budget, and industry. The threshold effect in the data makes this consequential rather than a matter of taste.
Semrush and Kevin Indig, analyzing 1,000 randomly selected domains across ChatGPT, ChatGPT with Search, Gemini, Google AI Overviews, and Perplexity, found Authority Score correlating with AI mentions at a Pearson coefficient of 0.65 and Spearman of 0.57 - but with AI share of voice at only 0.23 Pearson and 0.36 Spearman. Share of voice barely moved with incremental authority below a domain tier, responding only at higher bands.
Read against site situation, that produces three positions:
- New sites, minimal authority. Links below the threshold buy presence, not share. Pursue relevance and indexation over authority metrics, and expect flat returns until a base is established.
- Established sites below the tier. This is where acquisition work has the steepest slope. Concentrated, relevant links move both tracks.
- High-authority sites. Additional links produce diminishing ranking returns. Mention volume and topic-level coverage become the effective levers.
How to Evaluate a Link Opportunity Before Pursuing It
Evaluate an opportunity by checking indexation, topical relevance, page-level authority, outbound link density, and the linking page’s traffic - in that order, because a failure on any one makes the rest irrelevant.
The Metrics and What They Miss
Third-party authority metrics - Domain Rating, Domain Authority, Authority Score - estimate a site’s link profile strength. They correlate closely with each other and are reasonably reliable at what they measure.
What they miss is everything not in the link graph. A metric cannot tell you whether the page has readers, whether the site sells placements, whether the content is written for humans, or whether your subject appears anywhere on the domain. Sites optimizing to sell links optimize these exact metrics, which makes a high score on a placement-selling site an artefact rather than a signal.
Manual Qualification Signals
Six checks catch what metrics miss:
- Traffic - does the site receive organic visits, or only metric-inflating links?
- Publication pattern - regular editorial output, or dormancy punctuated by sponsored posts?
- Author attribution - real, identifiable writers with histories?
- Outbound profile - who else does the site link to, and in what categories?
- Placement disclosure - does the site advertise “write for us” with pricing?
- Topical fit - has the site covered your subject before?
A site failing three or more of these is a placement operation regardless of its authority score.
Outreach: What Determines Response Rate
Response rate is determined primarily by prospect qualification, secondarily by pitch relevance, and only marginally by pitch wording. Sending a better email to the wrong recipient does not work.
Prospect Qualification
Qualify prospects by whether they have a reason to link before you ask. A writer who has covered your topic, cited a competitor, or published a page with a gap your resource fills has a reason. A randomly selected site in your industry does not.
Qualification also determines volume. A qualified list of 40 contacts outperforms an unqualified list of 400, and does not train recipients to treat your domain as a spam source.
Pitch Construction
Effective pitches identify the specific page, state what you are offering in the first sentence, and explain the benefit to the recipient’s readers rather than to you. Length works against you: a pitch that takes more than fifteen seconds to parse is usually deleted.
Personalization means demonstrating you read the page. Inserting a first name into a template is not personalization, and recipients recognize the pattern.
Follow-Up Sequencing
How many follow-ups before stopping? Two, spaced roughly five and twelve days after the original message. Response rates drop steeply after the second, and additional messages generate complaints rather than links.
Each follow-up should add something - a different angle, a new asset, a relevant update - rather than repeating the original request.
Measuring Whether Link Building Worked
Measure link building on referring domains gained, rankings for target pages, organic traffic to those pages, and now citation frequency in AI answers. Link count alone measures activity, not outcome.
Leading Versus Lagging Indicators
Leading indicators move within weeks: outreach response rate, links placed, referring domains gained, referral traffic. These tell you whether the process is working.
Lagging indicators move over months: rankings for target keywords, organic sessions, conversions from organic, and citation frequency. These tell you whether the process produced value. Judging a program on lagging indicators before the lag has elapsed produces false negatives.
Attribution Lag: Why Effects Are Delayed
Link effects are delayed by the time required for Google to crawl the linking page, index it, and recalculate the destination’s position. Practical lag runs from several weeks to several months, longer for pages that are crawled infrequently.
The lag has a diagnostic consequence: a ranking change you observe today reflects link and engagement signals from a period well before today. Attributing this week’s movement to last week’s placements is usually wrong.
Tracking Citations Alongside Links
Track brand mention frequency and AI citation rate alongside link metrics, because the two-track model means link metrics alone cannot detect movement on the citation track.
Measure how often your brand is named in response to category and use-case prompts, sampled at a consistent interval across the major assistants. Track unlinked mentions as a distinct metric from links, since they contribute to the stronger citation signal in their own right.
What Link Building Costs
Link building costs are driven by labour, not by links. A single editorial link from an authoritative publication typically represents between four and fifteen hours of qualification, asset creation, outreach, and follow-up.
In-House Versus Agency Versus Freelance
In-house costs the loaded salary of the person doing it and produces the deepest subject knowledge, at the price of a slower start and no existing media relationships.
Agency costs a monthly retainer and supplies process, tooling, and relationships immediately. Quality varies enormously, and the variance tracks whether the agency does outreach or buys placements.
Freelance sits between the two on cost and varies most on outcome, since a freelancer’s results depend almost entirely on the individual’s existing relationships.
The determining question for all three is the same: does the provider earn links or purchase them? Ask for the acquisition method behind three recent links and check the resulting pages.
Why Cheap Link Packages Underperform
Packages priced per link at low rates underperform because the economics only work through placement purchase or automation. Genuine editorial outreach cannot be delivered profitably at those prices.
The result is links from sites that exist to sell links: high authority metrics, no traffic, no editorial standards, dense outbound profiles. Those links are devalued at best, and at worst form the deliberate pattern that manual actions require.
Link Building Tips That Compound
Some link building work produces returns that continue after the work stops. These are the compounding practices.
- Publish original data. A cited statistic attracts links for years without further effort.
- Convert mentions before chasing new links. Reclamation has the highest conversion rate of any outreach type.
- Build relationships with journalists, not link targets. A relationship produces repeated coverage; a placement produces one link.
- Maintain the assets that earn links. An outdated resource stops attracting citations.
- Track mentions as a first-class metric. What is not measured on the citation track is not managed there.
- Fix internal linking before buying external links. Authority already on the site is cheaper to redistribute than to acquire.
Frequently Asked Questions
Is link building still relevant in 2026?
Yes. Links remain a primary ranking signal for classic organic results, but they correlate weakly with AI citation at 0.218 against 0.664 for branded mentions. Both tracks need attention.
How many backlinks do I need to rank?
No fixed number exists. Requirements depend on competitor link profiles for the specific query, page-level relevance, and content quality. Comparing referring domains against the pages currently ranking gives a realistic target.
How long does link building take to show results?
Between two and six months for most sites. Google must crawl and index the linking page before recalculating rankings, and competitive queries take longer than low-competition ones.
Do nofollow links help SEO?
Yes, indirectly. Google treats nofollow as a hint rather than a directive, and nofollow links show correlation with AI mentions close to follow links. They also drive referral traffic and brand mentions.
Are paid backlinks against Google’s guidelines?
Buying links that pass ranking signals violates Google’s link spam policy. Paid placements are permitted when marked with the sponsored attribute, which prevents them from passing ranking value.
Should I disavow bad backlinks?
Only with an active manual action for unnatural links, or a known paid link campaign. Google devalues most spammy inbound links automatically, and unnecessary disavowal removes value.
Conclusion
Link building is the acquisition of external links to increase search visibility, and it now operates against two tracks rather than one. Links drive classic organic rankings. Mentions, co-occurrence, and topic-level coverage drive citation in AI-generated answers, where backlinks correlate at 0.218 against 0.664 for branded web mentions.
Three points carry the decision:
- Link value is determined by relevance, page-level authority, placement, and the linking page’s traffic - not by domain metrics alone.
- Acquisition methods vary by an order of magnitude in yield and risk. Original research and digital PR score best on both.
- Returns are non-linear. Below a domain authority tier, additional links buy presence rather than share of voice.
